Estimation Methods for Indonesian SSMM

122 However, a trade-off does exist. Although full information methods are more efficient, they are prone to specification errors and are also subject to higher computational burden, although this is not a problem with the present computer technology. In particular, if we misspecify one of the equations in the system and estimate the parameters using single equation methods, only the misspecified equation will be poorly estimated. If we employ system estimation techniques, however, the poor estimates for the misspecified equation may contaminate estimates for the other equations.

5.2.2 Estimation Methods for Indonesian SSMM

As far as SSMM’s are concerned, Batini and Haldane 1999 calibrated their small-scale macroeconomic model using the stylized facts found in the UK economy rather than estimating it. Beechey et al. 2000 estimated the equations in the RBA model individually by Ordinary Least Squares OLS. They argued that performing single equation least squares is the best choice since their cross-equation variance-covariance matrix for the estimated residuals show little cross-equation correlations. They also argued further that by doing so, it will curb mis-specification from spreading to one or more equations. de Freitas and Muinhos 1999 also estimated the behavioural equations in the SSMM of the Brazilian economy using the OLS method. In contrast, Arreaza et al. 2003, in their paper on a small macroeconomic model for the Venezuelan economy, estimated two of the equations in their model using OLS methods and calibrated one of them. 123 Valadkhani 2004 has pointed out that data availability in developing countries is a restrictive factor in doing sound macroeconomic modelling. This is very true in the case of Indonesia where the quality of data might be somewhere in between the undependable and the reliable due to time lags in gathering island-wide economic data and frequent revisions. Given this, the model-builder should use robust and simple methods, such as OLS, which are not too sensitive to the quality of data. Klein 1989 also argued that applying a method of joint estimation such as maximum likelihood should be avoided when the quality of data is mediocre and it is subject to frequent revisions. Furthermore, in estimating our Indonesian SSMM, OLS may be acceptable since we do not impose any cross-equation restrictions. Nevertheless, a simultaneous or system technique for estimating the SSMM equations cannot be dismissed immediately. The reason is that the Indonesian SSMM is constructed to depict and trace the relationships of key macroeconomic variables which are important for the Indonesian economy. The macroeconomic relationships among these variables—embodied in the IS, LM, and inflation equations—are simultaneous, so that joint estimation of the system is appropriate. A joint estimation method also ensures that all information, including restrictions on parameters, are taken into account in arriving at the final coefficient estimates, which is likely to make them more reliable. Provided that the contemporaneous errors have a joint normal distribution, the use of the FIML estimation method is therefore justified. In the following section, we present parameter estimates for the Indonesian SSMM based on both OLS and FIML methods. 124

5.3 SSMM Estimates